International Choice Modelling Conference, International Choice Modelling Conference 2017

Font Size: 
Using advanced choice models to study animal behaviour
Marek Giergiczny, Stephane Hess

Last modified: 28 March 2017

Abstract


Recent progress in positioning technology has facilitated the collection of large amounts of spatial data on animals. Satellite telemetry allows collection of accurate relocations less than 1 minute apart. Spatial data collected at such high frequency open up new scenarios because they contain important information about behaviour and decisions made by animals while moving through the environment. Studies using such fine-scale data and dealing with animal movement and habitat and resource selection can be used to answer fundamental ecological questions related to species distributions and diversity (Manly et al. 2002; Fortin et al., 2005), home range formation (Moorcroft et al. 2006), and can result in important management tools for identifying movement corridors (Chetkiewicz, 2006), key habitats (Squires, 2013), and responses to disturbance (Roever et al., 2010). These developments in positioning technology have led to new opportunities for investigating resource selection by animals but also new challenges related to the development of proper tools for the analysis of these large amounts of information. The two currently most prominent approaches in ecology are Resource Selection Functions (RSF) and Step-Selection Function (SSF).

Resource Selection Functions (RSFs) are used to model habitat selection by animals using data from GPS locations (Manly et al. 2002). A RSF is defined as any statistical model deployed to estimate the relative probability of selecting a resource unit versus alternative possible resource units, which in most applications to date has been logistic regression. RSFs have become a dominant tool in habitat selection studies. Another powerful modelling approach in ecology is the Step-Selection Function (SSF), which has been developed to estimate resource selection by animals moving through a landscape (Fortin et al., 2005). The main advantage of using an SSF rather than RSF is that SSFs may better model choices animals make as movement is included and as it constrains selection and availability (Johnson et al. 2008), which enables association of parameters of movement rules with landscape features.

The main contribution of our paper is to use the state of the art choice modelling approaches in  modelling RSF and SFF. We make use of GPS locations collected within the GLOBE project which aimed to study brown bear behaviour in Poland and Sweden. Within this research project, over 10 million GPS locations for 50 individual bears were collected over a period of 4 years. In our work we have made use of these data and built MNL, LCM and Mixed Logit (MMNL) at both individual and sample levels. We also incorporate a large amount of interactions with bear-specific characteristics such as age, gender and number of cubs.

We model the choice of destinations of bears on individual ‘journeys’. Differences arise depending on whether these relate to dispersal, such as a young male searching for a new territory, or foraging by an older bear with its own territory. We have conducted our analysis relying on different spatial resolution units (i.e. 25ha, 1km2 and 100km2). Our analysis indicates that the resolution of spatial covariates is crucial, and heterogeneity occurring at fine spatial scales can be masked if the resolution is too large, which is consistent with findings reported in other studies (Boyce 2006; Bowye 1996).

A variety of different characteristics are used to describe the alternatives, such as road density and building density, land cover area for different types (barren land, forest, shrub land etc), forest age, terrain ruggedness, area protection (national parks and reserves), and vegetation index.

In almost all applications RSF and SFF, coefficients are estimated with simple conditional logistic regressions. Our review indicates that advanced choice models remain largely overlooked in ecological studies. We are aware of only one application which used Mixed Logit to estimate RSF (Duchesne et al. 2010) and none to estimate SFF.

Our work shows that there is a substantial amount of inter-bear preference heterogeneity among studied animals. Our results clearly show that the current practice in ecological applications assuming that animals have similar behaviour, reflected in fixed preference parameters, is too restrictive. Moreover, the results of our study indicate that assuming constant preferences at the animal level is too restrictive too. For almost all bears in our sample, the estimation of a MMNL model at the individual level gives a statistically better fit than the fixed parameter MNL model, indicating that in addition to inter-bear heterogeneity, there is a substantial intra-bear preference heterogeneity too.

Our analysis shows that using more advanced discrete choice models gives a much deeper understanding of brown bear behaviour and yields much better predictions of SSF which is a very promising tool in ecology, wildlife management and conservation. We think that our study will propagate the use of more advanced choice models in ecology.

Conference registration is required in order to view papers.